Empirical Specification

1. Baseline models:

IV.3.1.1. Determinants of Attendance

To examine the determinants of the attendance including the effects of the UPE, I employ

Aprimaryij which is an outcome variable for attendance. The attendance variable is a binary

taking a value of 1if a child i of the household j attends primary school and zero otherwise. Therefore, I estimate the following model using OLS estimator at the child level aged 7 to 20.

25 Baseline model:

Aprimaryij = β0 + β1UPE +β2Xij + βiDagei + Ѵi+εij (1)

In equation (1) Aprimaryij is the attendance in primary school of an individual i in household j

and UPE is the treatment variable which indicates the effects of UPE on attendance in primary education. It takes value of 1if EICV2005/06 and zero if EICV2000/01. β1 is my coefficient of interest since it measures the changes in the likelihood to attend primary school due to the abolition of school fees. Dage is the age fixed effect while Ѵi is the regions fixed effects.X is a

vector of control variables such as dummies for gender of the child, transfers received, education of the householdhead, poverty index, income, distance to school, labor market and the existence of the parents. The reason why I include these factors in the attendance model is that they all affect the household’s decision to enroll their children in primary education.

Furthermore, to be able to answer my research questions about whether poor children benefited from the UPE policy, I extend model (1) by including an interaction term of UPE which is the product of the UPE with poverty index variable (which is a dummy variable). The interaction term allows the UPE effect on attendance for poor kids who were exposed to the abolition of the school fees in 2003. Thus, I estimate the following interaction regression model:

Aprimaryij = β0 + β1UPE +β2POVERTij + β3(UPE*POVERTij) + β4Xij + βiDagei + Ѵi +εij (2)

UPE*POVERTij is a vector of the interacted regressors while (β1+β3) is my coefficient of

interest in model (2) which measures the effects of the UPE on poor children to attend primary education.POVERTij is a vector of poverty index variable, measuring the effect of being poor on

attendance. X is the vector of the control variables while Dagei and Ѵi are age fixed effect and

region fixed effect respectively and εij is the error term.

IV.3.1.2. Delayed enrollment Beseline model:

Aprimaryij = β0 + β1UPE+ βiDagei+ β2Xij +Ѵi +εij (3)

Where UPE is the variable of interest and DAgei (Age fixed effects) is a dummy for age and takes value of 1 if individual has age n and zero otherwise and the same hold for all cohorts from

26

age of 7 to 20. Ѵ is the region fixed effects while X is a vector of control variables and the error term ε. It is worthy to mention that in order to avoid the problem of “dummy variable trap” due to perfect multicollinearity in the age dummy regressors. I arbitrarily omit the dummy age variable for 7 years old cohort.

As for the purpose to explore delayed enrollment after the introduction of the UPE, I use age fixed effect regression model by extending model (3) to interaction between two binary variables regression model by the product of the treatment variable(UPE) with each of the age dummy variable of my interest (14-20 years old). The reason is that, the official age to attend primary school in Rwanda is 7 to 13 years old. Therefore, for those who are still attending primary school and whose age is beyond 13 years, are considered as late enrolled cohorts in this study. The cohorts of 14 to 20 years are hence the group that interests my study to check whether there has been an increase of attendance among cohorts of age 14 to 20 years after the enactment of the UPE.

Thus, to identify this, I estimate the following regression model: Aprimaryij = β0 + β1UPE+ βiDagei + β

’

i (UPE* DAgei)+ β2Xij+Ѵi +εij (4)

In this regression, the coefficient of interest is (β1 + β’i) and measures the effect of the UPE on the

age of a child to attend primary education. The interacted regressors (UPE* DAgei) are restricted to individuals with age from 14 to 20, while Dagei is the age dummy variable and includes age

of 7-20 years old. βi measures the effect of each child to have a certain age on attendance while

β’i indicates additional effect of the UPE on the age of a child to attend primary education. The

comparison between the two coefficients βi and (β1 + β ’

i) indicates whether there is a link between

delayed enrollment and the introduction of the UPE in 2003. Ѵi is the region fixed effects . UPE is the treatment variable and εij is the error term.

IV.3.1.3. Determinants of Completion

Given that I have repeated cross-sectional survey data with two time periods EICV 2000/01 and EICV2005/06 just before and after the enactment of the UPE in 2003, the interaction regression model fits with the data being used for the case of completion in this study. The same holds as for the case of attendance. By combining the interaction regression with both age and region fixed effects, the effects of unobserved factors that differ from one group to another but which

27

are constant over time within cohorts and regions are eliminated. Analyzing the UPE effects on completion in primary school, I estimate the following baseline model:

Beseline model:

Cprimaryij = β0 + β1UPE +β2Xij + βiDagei + Ѵi +εij (5)

Where Cprimaryij is the completion in primary school of an individual i in household j and the UPE is a dummy variable and β1 is the coefficient of interest in my study, measuring the effect of the UPE on completion in primary school. Dagei and Ѵi are age and region fixed effect while Xij is a vector of control variables, including gender, poverty index, income dummies, distance to school, transfers received dummies, education dummies of the householdhead, labor market, as well as dummies for the existence of the parents in the household, to control for omitted variable bias.

I furthermore extend the baseline model to the interaction model with respect to poverty index variable. The idea is in fact to explore the effect that UPE program might have caused to poor children as they complete their primary education in Rwanda.

Cprimaryij = β0 + β1UPE + β2POVERTij +β3(UPE*POVERTij) +β4Xij + βiDagei + Ѵi +εij (6)

Where UPE*POVERTij is the interacted term while X is a vector of control variables and Dagei

and Ѵi are age and region fixed effects respectively. In this part I have two coefficients of

interest, β2 indicating the effects of being poor on completion in primary education and (β1+β3)

which measures the effects of the UPE on poor children to complete primary education. The comparison of the two coefficients enables to answer the question of whether UPE has benefited poor children while completing primary education in Rwanda.

o Variables

Dependent variables:

Aprimaryij: is referred to as a dummy dependent variable for attendance, taking value of 1 if

attending and 0 otherwise. This variable includes children of age from 7 to 20 from both groups with and without effect of the universal primary education program, since from the data point of

28

view, attendance is observed among these cohorts. The symbol i represents individual child and j is for the household.

Cprimaryij: stands for a binary dependent variable for the completion model and is restricted to

children of age 13, 14 and 15 who have been affected by the policy while completing their primary school. The pre-UPE and post-UPE data will be taken from the EICV2000/01 and EICV2005/06 respectively. The two outcome variables will comparatively generate very insightful information to answer the research questions of my study in section 3.i.

Explanatory variables:

Regarding the attendance regression, there are three key explanatory variables used for identification in the regression estimates. The treatment variable, interacted regressors and a vector of age fixed effects which captures children of age 7 to 20 but the cohorts of interest are those of 14 to 20 years old who are considered to be late in attending the school, since they are beyond the normal age to attend primary school in Rwanda. The age fixed effect enables to examine delayed enrollment such that the summation of the both coeffiecients for UPE dummy and interaction of the UPE dummy variable with each of the age dummy from 14 to 20 years old cohorts, is compared with the age dummy without interaction at each age level (14-20). The two coefficients are compared to find out whether there has been an increase in attendance after the UPE at age level. The same holds for the completion case.

The variable “level of poverty” is controlled to well identify the effects of UPE among poor children. More importantly, it takes 6 years to complete primary school in Rwanda and a child starts at 7 years old and finishes at 13 years old. To investigate the UPE effect on completion, this study takes into account children of 13, 14 and 15 years old since they should have benefited from UPE while completing their Primary school.

Control Variables:

X in both models, stands for a vector including all control variables such as: Distance to school,

dummies for income, poverty index, financial transfers received8, education of household head,

8 _{Financial transfers are provided by the government to poor household to incur their living cost including education}

29

existence of parents in the household, labor market (if worked for wages for the last 12 months) and gender of the child (see data part in section IV). Dagei and Ѵi age fixed effect and region fixed effect respectively.

There are various hindrances associated with children schooling in Rwanda and which are expected to negatively affect children’s schooling outcomes despite the abolition of school fees. In general, schooling outcomes of children depends on the presence of their parents. Many of the children lost one or all of their parents due to the genocide and war in 1994 and which still negatively impacts their education outcomes.

The interest of my study is not to check whether the orphans benefited from UPE but rather the explanatory variable of interest in this case is the poverty index. Thus, I interact the treatment variable(UPE) with poverty index variable within attendance and completion model in order to detect whether UPE increase the number of poor children who attended and completed primary school after the enactement of Free education policy in 2003.